Diogo Almeida, a former OpenAI researcher who worked on the development of ChatGPT, has introduced Jev through his startup TypeSafe AI. Unlike large language models, Jev doesn’t generate text. Instead, it produces calibrated probabilities that developers can use for clearly defined decisions. This is intended to make the model faster, cheaper, and less prone to hallucinations.
Developers are testing Jev for software automation, classification, model routing, and monitoring AI agents. Early tests suggest Jev can outperform larger models on certain tasks or replace them at significantly lower cost. Almeida expects falling costs to lead to broader AI integration into software, creating a more stable ecosystem of intelligent applications.